PhD Oral Exam - Saghar Vahidi, Information and Systems Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Wide-Area Monitoring, Protection, and Control (WAMPAC) systems depend on precise time synchronization---from the Global Navigation Satellite System (GNSS) or the Precision Time Protocol (PTP)---to align Phasor Measurement Unit (PMU) data across geographically distributed substations. This dependence introduces a critical vulnerability: manipulation of the time reference can propagate into synchrophasor-based monitoring and control applications. This thesis investigates the security of WAMPAC systems against Time Synchronization Attacks (TSAs), developing resilient control strategies, optimization-based attack formulations, and learning-based detection across two key applications: Wide-Area Damping Controller (WADC) and Voltage Stability Monitoring (VSM).

On the control side, the thesis develops observer-based WADCs that maintain stability under two complementary TSA manifestations. For First-Order TSAs (FO-TSAs) that cause intermittent denial of controller updates, a periodic sampled-data design provides LMI-based exponential stability guarantees under explicit attack-budget constraints. For Delay-Causing TSAs (DC-TSAs) that inject time-varying delays into the feedback loop, an event-triggered design leverages Looped-Lyapunov Functionals (LLFs) for less conservative stability conditions, with the system's delay margin analytically computed via the Guardian Map theorem and Rekasius substitution.

On the monitoring side, the thesis formulates stealthy TSAs that distort the Impedance Stability Index (ISI) while evading Largest Normalized Residual Bad Data Detection (LNR-BDD). We develop two optimization-based models: a single-level formulation that minimizes the perceived stability margin and a bi-level, operator-aware formulation that explicitly models corrective voltage-control responses and shows how adversaries can exploit them to further degrade the actual margin. To counter these threats, we train an explainable, topology-aware Graph Neural Network (GNN) detector on physics-informed data, with conditional Generative Adversarial Network (GAN) augmentation restricted to the training/validation development pool and a fully non-synthetic held-out test set. The GNN also provides node-level attribution cues for post-hoc localization and operator triage.

We evaluate the control-side contributions on the Kundur two-area and New England 39-bus systems, with OPAL-RT real-time simulations supporting the theoretical results. On the monitoring side, we evaluate the attack framework on the IEEE 9-, 30-, and 118-bus systems, while we evaluate the detector on the IEEE 30- and 118-bus systems, with an additional four-scenario PMU-placement sensitivity study.